Using Generative Adversarial Networks for Intrusion Detection in Cyber-Physical Systems
Abstract
Cyber-physical systems (CPS) are widely used in mission-critical systems in the Department of Defense and the U.S. Navy. They also form the backbone of national critical infrastructure. However, CPS technologies often sacrifice security in exchange for increased availability and efficiency, thus becoming prominent targets in cyber-warfare. This thesis explored machine learning to develop training examples for intrusion-detection systems on cyber-physical systems. We developed two generative adversarial network(GAN) models and assessed their ability to generate and detect anomalous traffic at the packet level. We tested two CPS datasets that included attacks that exploit commonly known vulnerabilities in Internet-of-Things networks and industrial control systems. The results confirmed that a GAN could improve the performance of intrusion-detection systems for detecting anomalous CPS traffic.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 2020
- Accession Number
- AD1126557
Entities
People
- Jessica L. Purser
Organizations
- Naval Postgraduate School